Mechanisms of aural encoding: I. Distinctive features for consonants!

R. A. COLE2, RALPH NORMAN HABER, AND B. D. SALES UNIVERSITY OF ROCHESTER

To examine the importance of distinctive features that are consonant is paired with the same vowel, while the used to encode consonants (following Wickelgren's analysis) height and slope of the pointed end of the bar repre­ in an immediate recall task, sequences of 5 consonants, all sent the transition cue by which consonants can be paired with the vowel/a/ were constructed and presented identified. This figure has been included here in aurally for recall. The middle three items in each sequence order to show the very close correspondence between all had either the same place of articulation (front, middle, acoustic parameters and articulation. Going across or back of the vocal apparatus), orthe same manner of articu­ any row, one can see that there is an identical acous­ lation (voic ed, unlJoic ed, or nasal), or were unrelated in tic parameter (the bottom formant) and an identical either place or manner (control). 1r was showl! that, in com­ manner of articulation. Going down any column. parison with the control sequences, consonants imbedded there is an identical acoustic parameter (the top formant) and an identical place of articulation. Since among others articulated similarly were recalled less accu­ there is such a close correspondence between the rately, suggesting that these distinctive features are impor­ acoustic and articulatory parameters, it is impossible tant in encoding and memory maintenance. A comparison of to devise stimuli which differ on only one of these the J manner and 3 place features showed that the greatest dimensions, while the other is held constant. difficulty in recall occurred for the similar manner sequences There are several possible approaches by which (especially voiced and unvoiced, implicating manner of to proceed in order to examine the encoding process. articulation as the critical distinctive feature in aural encod­ One converging operation has been to present white ing. Some discussion is also presented of a distinction noise to the S during rehearsal. White noise should between articulation and acoustic factors in encoding pro­ act to prevent S from "hearing" the sOW1d rehearsed, cesses. but should not interfere with articulatory movements. Studies measuring errors in short-term memory When this condition has been employed (e.g., Hintzman, have fOW1d that stimuli which sOW1d alike are more 1965; Murray, 1965), white noise does not interfere apt to be confused with each other in a serial re­ with performance. While this result supports a stress call task. While these experiments show the impor­ on an articulatory mechanism of encoding, sperling tance of acoustic factors in short-term memory, (1963) reports that Ss hear the noise outside of they cannot pinpoint the mechanism of encoding since their head, while rehearsal was going on inside of the usual stimuli used have both acoustic and artic­ their head. Thus, white noise may be an ambiguous ulatory features in common. Thus, in a string such or ineffective converging operation, since S may as B D C T P G, the letters all have the sOW1d "E," PLACE OF ARTICULATION and the tongue always glides toward the same position. FRONT MIDDLE BACK There can either be an acoustic representation of the stimulation in the nervous system, or an encoding from feedback of articulatory movements, or some \- combination of these. r Psychologists have tended to treat this distinction J ],a. _I" cla _I sa in an "either-or" manner, assuming either an acoustic 0;;. or articulatory mechanism of aural encoding. It has :, ., long been known, however, that there is a close ....i - correspondence between the articulation of a sOW1d i' /# and the acoustic makeup of that sOW1d. The invention pa. ta ka of the sOW1d spectograph has enabled acoustic pho­ neticians to measure precisely the acoustic parameters \", \- of a sOW1d. In Fig. I, taken from Liberman (1957), =r - - stimuli in the same row have an identical manner of - rna na - na articulation, while stimuli in the same column have - an identical place of articulation. The dark bars in Fig. 1. Specwgraphic patterns that illustrate the transition cues for a stop and nasal consonants in initial position with the vowel each box are concentrations of energy at certain / a/. The dotted portions in the second row indicate the presence frequencies representing vowel resonance. Theheightof of noise (aspiration) in place of hannonics (taken from Libennan, the "body" of each bar is identical because each 1957).

Perception & Psychophysics, 1968, Vol. 3 (48) Copyright 1968. Psychonomic Journals, Santa Barbara, Cali], 281 handle it as an ignored channel (e.g., Broadbent, to each S seated in a small room with the E. They 1958). It should be possible to prevent S from ig­ were recorded by the first author who has had noring it, such as by requiring him to monitor the training in linguistics and in pronunciation of phones. amplitude of the white noise by means of a pointer They were recorded so that they were heard at a while he is encoding and rehearsing the material rate of one per second. Seven different types of se­ for recall. If recall was still unimpaired, the artic­ quences were used, differing according to the rules ulatory as compared to the acoustic factor would be of selection of the items from Fig. 1. In the three further indicated as the rehearsal mechanism. different manner of articulation conditions, the second A reverse procedure, one to permit acoustic en­ third, and fourth sound in each sequence was the coding but not the articulatory responses, might be same manner of articulation. Looking at Fig. 1 this to prevent S from rehearsing, but during the encoding means that the stimuli from the same row were in and rehearsal time, have S hear a rehearsal in his these positions. The sounds on either side of the earphones. He could be prevented from rehearsing sequence had neither place nor manner similar to by assigning him some other task during the period. the adjacent sound. Such a sequence for voiced con­ This type of design is replete with problems, and sonants would be TA, BA, DA, GA, NA. In the three to our knowledge has not been tried yet, but it does different place of articulation conditions the second, represent a possibility. third, and fourth stimuli in each sequence had the A third approach does not attempt to converge same place of articulation. In reference to Fig. I, directly on a test between articulatory or acoustic this means that the stimuli were from the same mechanisms of encoding. As an example, Wickelgren column. The end positions were neither place nor (1966) studied intrusion errors for English consonants manner of articulation similar to the adjacent sounds. when all such stimuli were paired with the same Such a sequence would be NA, BA, PA, MA, GA. In vowel sound. He found that intrusion errors could the control condition, no sound in any of the positions be accurately predicted from a linguistic distinctive had either place nor manner of articulation similar feature system in terms of voicing, nasality, place to adjacent sounds. Such a sequence would be BA, of articulation, and openess of the vocal tract. His TA, GA, NA, PA. For each of the seven conditions results suggest that sounds are encoded into short­ six sequences were used, by constructing six sets term memory as a set of specific features, and a of five sounds meeting these rules (except for the forgotten stimulus will be replaced by one with the control condition, which had 18 sequences). Thus, most similar features. each Sheard 54 sequences. Within each condition The present experiment employs a related approach. of six sequences the sounds were balanced over If Wickelgren is correct, then it should be possible positions. No sound was repeated within a sequence. to construct sequences of consonants for recall in Each S received the 54 sequences in the same order, which specific features can be embedded in contexts which was constructed from a random selection of of differing amounts of confusability. Thus, if voicing the seven conditions. is an important feature in encoding and rehearsal, Ss should have great difficulty encoding a sequence Procedure In which all of the consonants are voiced. By com­ The S was instructed that he would hear a list paring sequences which differ in the feature held in of five sounds. After the last one he should wait common, it should be possible to order the relative 3 sec and then report them in the order he heard importance of these features as used in encoding them. During the report he was told that he would and rehearsal. have as much time as he needed. He was told to The stimuli were chosen from Fig. 1. Since the rehearse the sound as he heard it and not just try columns differed only in the place of articulation, to remember the consonant letter. The Ss' responses the rows only in the manner of articulation, sequences were tape recorded. For the basic analyses, the S were constructed that placed consonants together from was scored as correct for each sound in a sequence the same row, or the same column. only if he named the proper sound in the proper position. METHOD Subjects RESUL TS Fifty undergraduate University of Rochester stu­ Figure 2 shows the accuracy of recall of the nine dents served as volunteer Ss. They were tested sounds for the control sequences, in which any inter­ individually in a session lasting less than one hour. ference due to similarity should be minimized. The differences observed are statistically significant (p Stimuli < .001). A series of sequences of five consonants, each To determine the effects of similarity on recall, of which was paired with the vowel "a," were se­ each sound in each of the three middle positions lected from among the nine consonants in Table 1. and in each condition was compared to the same They were presented hy a Wollensak tape recorder sound in the respective position of the control se-

Perceptum & Psychophysics, 1968, Vol. 3 (48) 80 Percent Correct For Control Condition fered from the control in position two; none did in position four, while all three of the manner sequences

NG did in position three (voiced, voiceless, and nasal). These latter comparisons suggested an interaction between position and condition which in fact was 70 significant in the analysis of variance (p < .01). A main effect of condition (p< .05) indicated that the three manner conditions were generally more affected in all three positions by similarity of context than

p were the three place conditions. This was also sup­ 60 ported by the pattern of t test findings.

N DISCUSSION This experiment demonstrates: (1) it is possible to create interference in short-term memory by 50 embedding sounds with similar distinctive features in the same context; (2) the context effect is greatest for the middle sounds, with little or no effect for the end sounds in the context; and (3) contexts that have similar manner of articulation create greater K Voiced Manner Unvoiced Manner Nosal Manner interference than contexts of similar place of artic­ L 40 ---====='----_====='=-_----'=====___ ulation. Fig. 2. Percentage of correct recall responses for each of the The interference effect shown in this study gives nine sounds used in the control sequences. Each point is summed over fifty Ss, and for each S, is summed over all positions. a powerful support to Wickelgren's notion that sounds are encoded in short-term memory as a set of quences. The percentage correct for each sound was features. It has now been shown that a distinctive determined and then subtracted from the comparable feature hypothesis accurately predicts intrusion er­ score from the control sequences. Thus, for a partic­ rors in short-term memory, and features show sig­ ular sound in a particular condition, a positive number nificant interference when sounds with similar features implies that the sound was easier to recall in the are embedded in the same context. The interference control condition than in a context of similar sounds. effect when similar features are used is a more The scores for the three sounds defining. each con­ powerful demonstration of encoding mechanisms than dition were combined, and these are shown in Fig. 3. confusions in gross strings such as B CPT D in An analysis of variance showed a main effect for which one may phenomenally identify the "E" sound position (p< .001), indicating that position three pro­ in all stimuli. In sequences used in this experiment, duced the greatest differences between the control one need be familiar with phonetics in order to and the six experimental sequences. Individual t tests point out the similar features. indicated that only the nasal manner sequences dif- The amount of interference seems unrelated to the overall ease of recalling the sound in a nonconfusable 20 context. Thus in Fig. 1 (the control sequences) MA ~ 18 0 NA and NGA are easier to recall than BA DA and u 16 ~ GA, but there is more interference created when 14 MA NA and NGA are placed together in the same § 12 ~ sequence. 0 u 10 It appears as if distinctive features involving man­ s: 0 8 0 ner of articulation (voicing, lack of voicing, nasality)

Perception & Psychophysics, 1968, \'01. 3 (4B) 283 Hintzman, D. L. Classification and aural coding in short-term States Public Health Service MH 10753, and by a grant from the memory. Psuctum, Sci., 1965. 3, 161-162. National Science Foundation GB 2909, to the second author. and by Libennan, A. M. Some results of research on speech perception. the Undergraduate Research Participation Program grant to the J. Acoust. Soc. Amer., 1957,29, 117-123. Psychology Department of the University of Rochester from the Murray, D. J. The effect of white noise upon the recall of vocalized National Science Foundation GY 2619 in an award to the first lists. Canad. J. Psycho/., 1965, 19, 333-345. author. We would like to thank Barbara Herr for her assistance in Sperling, G. A model for visual memory tasks. Hum. Factors. data collection and analysis, Audrey Konow for computer pro­ 1963,5, 17-31. gramming, and Robert Kulikowski for his participation in running Wickelgren, W. A. Distinctive features and errors in short-term the experiment. memory for English consonants. J. Acoust. Soc. Amer., 1966, 2. Now at the University of California at Riverside. 39, 388-398. Notes 1. This research was supported in part by a grant from the United (Accepted for publication October 9, 1967.)

Abstract: Structure in perceptual classification by S. lmai, Hokkaido University, and W. R. Garner, Yale University

An article was recently published as a Psychonomic the total set so as to include different stimuli specified by Monograph Supplement that should be of interest to E. This experiment showed that the structure of the subset readers of this Journal. It was published as a Mono­ itself and of the total set both influenced how subsets were graph because it was a long article and because selected. The third experiment required selections of two space in this Journal is under heavy pressure, but subsets, again so as to include different stimuli specified an abstract of it is given below so that readers by E. This experiment showed that the structure of the two will know of its existence and may either purchase subsets combined is similar to the structure of selected it (price $ 1.00) or refer to it in their libraries. single subsets. The fourth experiment required complete Imai, S. and Garner, W. R. Structure in perceptual classification (partitioning) of subsets of 4, 5, or 6 different classification. Pyschon. Monogr. Supplv, 1968, 2, No. stimuli selected by E. Regardless of the size or structure of 9 (Whole No. 25): a subset, classification tended to maintain the attribute Four experiments were carried out on the role of structure structure of the total set. Taken together, these experiments of stimulus sets in perceptual classification. All experiments demonstrate that attribute structure of the total set and sub­ used a total set of nine visual stimuli derived from three sets is more important in determining perceptual classifica­ forms times three colors. The first experiment showed that tion than simple similarity and difference relations between the judged similarity of any pair of stimuli depended only on stimuli. Furthermore, there is the strong suggestion that whether the pair differed by one or by two attributes; thus subsets or classes are perceived more in relation to larger there was in effect no metric structure of the attributes. The classes and the total set than in contrast to other classes of second experiment required S to select a single subset from the same size.

Perception & Psychophysics. 1968, Vol. 3 ('18)